ds.c

linux 内核源代码

/* ds.c: Domain Services driver for Logical Domains
 *
 * Copyright (C) 2007 David S. Miller <[e-mail]>
 */

#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/types.h>
#include <linux/string.h>
#include <linux/slab.h>
#include <linux/sched.h>
#include <linux/delay.h>
#include <linux/mutex.h>
#include <linux/kthread.h>
#include <linux/reboot.h>
#include <linux/cpu.h>

#include <asm/ldc.h>
#include <asm/vio.h>
#include <asm/mdesc.h>
#include <asm/head.h>
#include <asm/irq.h>

#define DRV_MODULE_NAME		"ds"
#define PFX DRV_MODULE_NAME	": "
#define DRV_MODULE_VERSION	"1.0"
#define DRV_MODULE_RELDATE	"Jul 11, 2007"

static char version[] __devinitdata =
	DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
MODULE_AUTHOR("David S. Miller ([e-mail])");
MODULE_DESCRIPTION("Sun LDOM domain services driver");
MODULE_LICENSE("GPL");
MODULE_VERSION(DRV_MODULE_VERSION);

struct ds_msg_tag {
	__u32			type;
#define DS_INIT_REQ		0x00
#define DS_INIT_ACK		0x01
#define DS_INIT_NACK		0x02
#define DS_REG_REQ		0x03
#define DS_REG_ACK		0x04
#define DS_REG_NACK		0x05
#define DS_UNREG_REQ		0x06
#define DS_UNREG_ACK		0x07
#define DS_UNREG_NACK		0x08
#define DS_DATA			0x09
#define DS_NACK			0x0a

	__u32			len;
};

/* Result codes */
#define DS_OK			0x00
#define DS_REG_VER_NACK		0x01
#define DS_REG_DUP		0x02
#define DS_INV_HDL		0x03
#define DS_TYPE_UNKNOWN		0x04

struct ds_version {
	__u16			major;
	__u16			minor;
};

struct ds_ver_req {
	struct ds_msg_tag	tag;
	struct ds_version	ver;
};

struct ds_ver_ack {
	struct ds_msg_tag	tag;
	__u16			minor;
};

struct ds_ver_nack {
	struct ds_msg_tag	tag;
	__u16			major;
};

struct ds_reg_req {
	struct ds_msg_tag	tag;
	__u64			handle;
	__u16			major;
	__u16			minor;
	char			svc_id[0];
};

struct ds_reg_ack {
	struct ds_msg_tag	tag;
	__u64			handle;
	__u16			minor;
};

struct ds_reg_nack {
	struct ds_msg_tag	tag;
	__u64			handle;
	__u16			major;
};

struct ds_unreg_req {
	struct ds_msg_tag	tag;
	__u64			handle;
};

struct ds_unreg_ack {
	struct ds_msg_tag	tag;
	__u64			handle;
};

struct ds_unreg_nack {
	struct ds_msg_tag	tag;
	__u64			handle;
};

struct ds_data {
	struct ds_msg_tag	tag;
	__u64			handle;
};

struct ds_data_nack {
	struct ds_msg_tag	tag;
	__u64			handle;
	__u64			result;
};

struct ds_info;
struct ds_cap_state {
	__u64			handle;

	void			(*data)(struct ds_info *dp,
					struct ds_cap_state *cp,
					void *buf, int len);

	const char		*service_id;

	u8			state;
#define CAP_STATE_UNKNOWN	0x00
#define CAP_STATE_REG_SENT	0x01
#define CAP_STATE_REGISTERED	0x02
};

static void md_update_data(struct ds_info *dp, struct ds_cap_state *cp,
			   void *buf, int len);
static void domain_shutdown_data(struct ds_info *dp,
				 struct ds_cap_state *cp,
				 void *buf, int len);
static void domain_panic_data(struct ds_info *dp,
			      struct ds_cap_state *cp,
			      void *buf, int len);
#ifdef CONFIG_HOTPLUG_CPU
static void dr_cpu_data(struct ds_info *dp,
			struct ds_cap_state *cp,
			void *buf, int len);
#endif
static void ds_pri_data(struct ds_info *dp,
			struct ds_cap_state *cp,
			void *buf, int len);
static void ds_var_data(struct ds_info *dp,
			struct ds_cap_state *cp,
			void *buf, int len);

struct ds_cap_state ds_states_template[] = {
	{
		.service_id	= "md-update",
		.data		= md_update_data,
	},
	{
		.service_id	= "domain-shutdown",
		.data		= domain_shutdown_data,
	},
	{
		.service_id	= "domain-panic",
		.data		= domain_panic_data,
	},
#ifdef CONFIG_HOTPLUG_CPU
	{
		.service_id	= "dr-cpu",
		.data		= dr_cpu_data,
	},
#endif
	{
		.service_id	= "pri",
		.data		= ds_pri_data,
	},
	{
		.service_id	= "var-config",
		.data		= ds_var_data,
	},
	{
		.service_id	= "var-config-backup",
		.data		= ds_var_data,
	},
};

static DEFINE_SPINLOCK(ds_lock);

struct ds_info {
	struct ldc_channel	*lp;
	u8			hs_state;
#define DS_HS_START		0x01
#define DS_HS_DONE		0x02

	u64			id;

	void			*rcv_buf;
	int			rcv_buf_len;

	struct ds_cap_state	*ds_states;
	int			num_ds_states;

	struct ds_info		*next;
};

static struct ds_info *ds_info_list;

static struct ds_cap_state *find_cap(struct ds_info *dp, u64 handle)
{
	unsigned int index = handle >> 32;

	if (index >= dp->num_ds_states)
		return NULL;
	return &dp->ds_states[index];
}

static struct ds_cap_state *find_cap_by_string(struct ds_info *dp,
					       const char *name)
{
	int i;

	for (i = 0; i < dp->num_ds_states; i++) {
		if (strcmp(dp->ds_states[i].service_id, name))
			continue;

		return &dp->ds_states[i];
	}
	return NULL;
}

static int __ds_send(struct ldc_channel *lp, void *data, int len)
{
	int err, limit = 1000;

	err = -EINVAL;
	while (limit-- > 0) {
		err = ldc_write(lp, data, len);
		if (!err || (err != -EAGAIN))
			break;
		udelay(1);
	}

	return err;
}

static int ds_send(struct ldc_channel *lp, void *data, int len)
{
	unsigned long flags;
	int err;

	spin_lock_irqsave(&ds_lock, flags);
	err = __ds_send(lp, data, len);
	spin_unlock_irqrestore(&ds_lock, flags);

	return err;
}

struct ds_md_update_req {
	__u64				req_num;
};

struct ds_md_update_res {
	__u64				req_num;
	__u32				result;
};

static void md_update_data(struct ds_info *dp,
			   struct ds_cap_state *cp,
			   void *buf, int len)
{
	struct ldc_channel *lp = dp->lp;
	struct ds_data *dpkt = buf;
	struct ds_md_update_req *rp;
	struct {
		struct ds_data		data;
		struct ds_md_update_res	res;
	} pkt;

	rp = (struct ds_md_update_req *) (dpkt + 1);

	printk(KERN_INFO "ds-%lu: Machine description update.\n", dp->id);

	mdesc_update();

	memset(&pkt, 0, sizeof(pkt));
	pkt.data.tag.type = DS_DATA;
	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
	pkt.data.handle = cp->handle;
	pkt.res.req_num = rp->req_num;
	pkt.res.result = DS_OK;

	ds_send(lp, &pkt, sizeof(pkt));
}

struct ds_shutdown_req {
	__u64				req_num;
	__u32				ms_delay;
};

struct ds_shutdown_res {
	__u64				req_num;
	__u32				result;
	char				reason[1];
};

static void domain_shutdown_data(struct ds_info *dp,
				 struct ds_cap_state *cp,
				 void *buf, int len)
{
	struct ldc_channel *lp = dp->lp;
	struct ds_data *dpkt = buf;
	struct ds_shutdown_req *rp;
	struct {
		struct ds_data		data;
		struct ds_shutdown_res	res;
	} pkt;

	rp = (struct ds_shutdown_req *) (dpkt + 1);

	printk(KERN_ALERT "ds-%lu: Shutdown request from "
	       "LDOM manager received.\n", dp->id);

	memset(&pkt, 0, sizeof(pkt));
	pkt.data.tag.type = DS_DATA;
	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
	pkt.data.handle = cp->handle;
	pkt.res.req_num = rp->req_num;
	pkt.res.result = DS_OK;
	pkt.res.reason[0] = 0;

	ds_send(lp, &pkt, sizeof(pkt));

	orderly_poweroff(true);
}

struct ds_panic_req {
	__u64				req_num;
};

struct ds_panic_res {
	__u64				req_num;
	__u32				result;
	char				reason[1];
};

static void domain_panic_data(struct ds_info *dp,
			      struct ds_cap_state *cp,
			      void *buf, int len)
{
	struct ldc_channel *lp = dp->lp;
	struct ds_data *dpkt = buf;
	struct ds_panic_req *rp;
	struct {
		struct ds_data		data;
		struct ds_panic_res	res;
	} pkt;

	rp = (struct ds_panic_req *) (dpkt + 1);

	printk(KERN_ALERT "ds-%lu: Panic request from "
	       "LDOM manager received.\n", dp->id);

	memset(&pkt, 0, sizeof(pkt));
	pkt.data.tag.type = DS_DATA;
	pkt.data.tag.len = sizeof(pkt) - sizeof(struct ds_msg_tag);
	pkt.data.handle = cp->handle;
	pkt.res.req_num = rp->req_num;
	pkt.res.result = DS_OK;
	pkt.res.reason[0] = 0;

	ds_send(lp, &pkt, sizeof(pkt));

	panic("PANIC requested by LDOM manager.");
}

#ifdef CONFIG_HOTPLUG_CPU
struct dr_cpu_tag {
	__u64				req_num;
	__u32				type;
#define DR_CPU_CONFIGURE		0x43
#define DR_CPU_UNCONFIGURE		0x55
#define DR_CPU_FORCE_UNCONFIGURE	0x46
#define DR_CPU_STATUS			0x53

/* Responses */
#define DR_CPU_OK			0x6f
#define DR_CPU_ERROR			0x65

	__u32				num_records;
};

struct dr_cpu_resp_entry {
	__u32				cpu;
	__u32				result;
#define DR_CPU_RES_OK			0x00
#define DR_CPU_RES_FAILURE		0x01
#define DR_CPU_RES_BLOCKED		0x02
#define DR_CPU_RES_CPU_NOT_RESPONDING	0x03
#define DR_CPU_RES_NOT_IN_MD		0x04

	__u32				stat;
#define DR_CPU_STAT_NOT_PRESENT		0x00
#define DR_CPU_STAT_UNCONFIGURED	0x01
#define DR_CPU_STAT_CONFIGURED		0x02

	__u32				str_off;
};

static void __dr_cpu_send_error(struct ds_info *dp,
				struct ds_cap_state *cp,
				struct ds_data *data)
{
	struct dr_cpu_tag *tag = (struct dr_cpu_tag *) (data + 1);
	struct {
		struct ds_data		data;
		struct dr_cpu_tag	tag;
	} pkt;
	int msg_len;

	memset(&pkt, 0, sizeof(pkt));
	pkt.data.tag.type = DS_DATA;
	pkt.data.handle = cp->handle;
	pkt.tag.req_num = tag->req_num;
	pkt.tag.type = DR_CPU_ERROR;
	pkt.tag.num_records = 0;

	msg_len = (sizeof(struct ds_data) +
		   sizeof(struct dr_cpu_tag));

	pkt.data.tag.len = msg_len - sizeof(struct ds_msg_tag);

	__ds_send(dp->lp, &pkt, msg_len);
}

static void dr_cpu_send_error(struct ds_info *dp,
			      struct ds_cap_state *cp,
			      struct ds_data *data)
{
	unsigned long flags;

	spin_lock_irqsave(&ds_lock, flags);
	__dr_cpu_send_error(dp, cp, data);
	spin_unlock_irqrestore(&ds_lock, flags);
}

#define CPU_SENTINEL	0xffffffff

static void purge_dups(u32 *list, u32 num_ents)
{
	unsigned int i;

	for (i = 0; i < num_ents; i++) {
		u32 cpu = list[i];
		unsigned int j;

		if (cpu == CPU_SENTINEL)
			continue;

		for (j = i + 1; j < num_ents; j++) {
			if (list[j] == cpu)
				list[j] = CPU_SENTINEL;
		}
	}
}

static int dr_cpu_size_response(int ncpus)
{
	return (sizeof(struct ds_data) +
		sizeof(struct dr_cpu_tag) +
		(sizeof(struct dr_cpu_resp_entry) * ncpus));
}

static void dr_cpu_init_response(struct ds_data *resp, u64 req_num,
				 u64 handle, int resp_len, int ncpus,
				 cpumask_t *mask, u32 default_stat)
{
	struct dr_cpu_resp_entry *ent;
	struct dr_cpu_tag *tag;
	int i, cpu;

	tag = (struct dr_cpu_tag *) (resp + 1);
	ent = (struct dr_cpu_resp_entry *) (tag + 1);

	resp->tag.type = DS_DATA;
	resp->tag.len = resp_len - sizeof(struct ds_msg_tag);
	resp->handle = handle;
	tag->req_num = req_num;
	tag->type = DR_CPU_OK;
	tag->num_records = ncpus;

	i = 0;
	for_each_cpu_mask(cpu, *mask) {
		ent[i].cpu = cpu;
		ent[i].result = DR_CPU_RES_OK;
		ent[i].stat = default_stat;
		i++;
	}
	BUG_ON(i != ncpus);
}

static void dr_cpu_mark(struct ds_data *resp, int cpu, int ncpus,
			u32 res, u32 stat)
{
	struct dr_cpu_resp_entry *ent;
	struct dr_cpu_tag *tag;
	int i;

	tag = (struct dr_cpu_tag *) (resp + 1);
	ent = (struct dr_cpu_resp_entry *) (tag + 1);

	for (i = 0; i < ncpus; i++) {
		if (ent[i].cpu != cpu)
			continue;
		ent[i].result = res;
		ent[i].stat = stat;
		break;
	}
}

static int dr_cpu_configure(struct ds_info *dp,
			    struct ds_cap_state *cp,
			    u64 req_num,
			    cpumask_t *mask)
{
	struct ds_data *resp;
	int resp_len, ncpus, cpu;
	unsigned long flags;

	ncpus = cpus_weight(*mask);
	resp_len = dr_cpu_size_response(ncpus);
	resp = kzalloc(resp_len, GFP_KERNEL);
	if (!resp)
		return -ENOMEM;

	dr_cpu_init_response(resp, req_num, cp->handle,
			     resp_len, ncpus, mask,
			     DR_CPU_STAT_CONFIGURED);

	mdesc_fill_in_cpu_data(*mask);

	for_each_cpu_mask(cpu, *mask) {
		int err;

		printk(KERN_INFO "ds-%lu: Starting cpu %d...\n",
		       dp->id, cpu);
		err = cpu_up(cpu);
		if (err) {
			__u32 res = DR_CPU_RES_FAILURE;
			__u32 stat = DR_CPU_STAT_UNCONFIGURED;

			if (!cpu_present(cpu)) {
				/* CPU not present in MD */
				res = DR_CPU_RES_NOT_IN_MD;
				stat = DR_CPU_STAT_NOT_PRESENT;
			} else if (err == -ENODEV) {
				/* CPU did not call in successfully */
				res = DR_CPU_RES_CPU_NOT_RESPONDING;
			}

			printk(KERN_INFO "ds-%lu: CPU startup failed err=%d\n",
			       dp->id, err);
			dr_cpu_mark(resp, cpu, ncpus, res, stat);
		}
	}

	spin_lock_irqsave(&ds_lock, flags);
	__ds_send(dp->lp, resp, resp_len);
	spin_unlock_irqrestore(&ds_lock, flags);

	kfree(resp);

	/* Redistribute IRQs, taking into account the new cpus.  */
	fixup_irqs();

	return 0;
}

static int dr_cpu_unconfigure(struct ds_info *dp,
			      struct ds_cap_state *cp,
			      u64 req_num,
			      cpumask_t *mask)
{
	struct ds_data *resp;
	int resp_len, ncpus, cpu;
	unsigned long flags;

	ncpus = cpus_weight(*mask);
	resp_len = dr_cpu_size_response(ncpus);
	resp = kzalloc(resp_len, GFP_KERNEL);
	if (!resp)
		return -ENOMEM;

	dr_cpu_init_response(resp, req_num, cp->handle,
			     resp_len, ncpus, mask,
			     DR_CPU_STAT_UNCONFIGURED);

	for_each_cpu_mask(cpu, *mask) {
		int err;

		printk(KERN_INFO "ds-%lu: Shutting down cpu %d...\n",
		       dp->id, cpu);
		err = cpu_down(cpu);
		if (err)
			dr_cpu_mark(resp, cpu, ncpus,
				    DR_CPU_RES_FAILURE,
				    DR_CPU_STAT_CONFIGURED);
	}

	spin_lock_irqsave(&ds_lock, flags);
	__ds_send(dp->lp, resp, resp_len);
	spin_unlock_irqrestore(&ds_lock, flags);

	kfree(resp);

	return 0;